# DeepMind Just Made The Most Powerful Game AI Engine! ## Метаданные - **Канал:** Two Minute Papers - **YouTube:** https://www.youtube.com/watch?v=YvuEKrJhjos - **Дата:** 17.08.2025 - **Длительность:** 6:42 - **Просмотры:** 135,258 - **Источник:** https://ekstraktznaniy.ru/video/12169 ## Описание ❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambda.ai/papers Guide: Rent one of their GPU's with over 16GB of VRAM Open a terminal Just get Ollama with this command - https://ollama.com/download/linux Then run ollama run gpt-oss:120b - https://ollama.com/library/gpt-oss:120b Genie 3: https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/ Sources: https://x.com/amoufarek/status/1955776162447102238 https://x.com/amoufarek/status/1955299375548076382 https://x.com/holynski_/status/1953882726656094622 https://x.com/holynski_/status/1953879983535141043 https://x.com/RuiHuang_art/status/1954716703340048877 https://x.com/mattmcgill_/status/1953827141700772186 📝 My paper on simulations that look almost like reality is available for free here: https://rdcu.be/cWPfD Or this is the orig. Nature Physics link with clickable citations: https://www.nature.com/articles/s41567-022-01788-5 🙏 We would like to thank our generous Patreon supporters who ma ## Транскрипт ### Segment 1 (00:00 - 05:00) [] This is Deep Mind's Genie 3 AI making video games. And it's a bit like handing Barb Ross a napkin and jumping into his masterpiece. And I have to be honest, I did not expect to make a video on this, but something happened. One, you fellow scholars asked for it vigorously. And two, this is going to happen. And I got to tell you about it. You have to know. Let me explain. Genie3 can take a text prompt from you and create a video game world out of it that you can control and in some cases you can even interact with it. So I am very happy to call them actual games. It is a follow-up work from Genie 2 and it is incredible that it got this much better in just 8 months. Yes, not 8 years. What you are seeing is just 8 months of progress but it can do so much more. For instance, here you see a video of two people using it to play a video game. Wait a minute. What? This is not a video. This is a video game within a video game. Genie 3 can do that, too. Wow. And check this out. Third person view, but not only that, but with proper camera movement. This is a level of consistency I have not seen anywhere yet. Wow. In the research demo, it also seems that they can give it an input image and then explore the world around it to your liking. And what if we do that with a classical painting? Are you saying that? Yes, that is exactly what I'm saying. You can give it the death of Socrates and make the painting come alive or just take a completely different style of painting and it works on that too. And at this point, I think you fellow scholars are seeing that this is going way beyond just creating games. For instance, you can give it your own artwork. Just dream up your own little world. Give it to the AI. And my words kind of fail me to describe how cool it is that it helps you unleash your creativity. Dear fellow scholars, this is two minute papers with Dr. Koa Eher, Dr. Carol John Alpha. And you can create something cool with text to image. And if you use a video generator AI, yes, you can extend it to a video, but you often get only 8 seconds worth of footage. But with Genie3, not anymore. You can extend that video to any length you desire. And don't forget, you choose your own camera angles. This is a video game world after all. But that's not even the best part yet. So now hold on to your papers fellow scholars because here comes the bombshell what I showed you at the start. It is called Sema. This is their AI that is able to learn to play these amazing 3D video games. Aha. I think you know what's coming. Oh yes. One AI dreams up new worlds and we put another AI in there to learn and play in them. And then you can dream up a new one and learn in that one too. So, two crazy things here. One, when training robots in a simulation, we often use a technique called domain randomization. This means that it learns to cook not just in one kitchen, but in 10,000 different kinds of kitchens. So, by the time it enters the real world, it really knows how to handle any scenario. But this with Genie 3, this is even better than domain randomization. It does not just randomize kitchens. It creates 10,000 whole new worlds to play in. Imagine what that could teach the AI systems of the future. Wow. But wait, there is an indication as to what they could learn already. And what drives me crazy is that absolutely nobody is talking about this. Here goes. When talking about the video game playing agent Sema, I did not expect this at all. You take a specialist AI and let it master one game. Then you take a generalist AI that kind of knows this game but knows other games too. So which one is better at this one game? Of course I say the specialist because it had played the game so much more. Just imagine the Olympics. You have a super muscular wrestler who did wrestling his whole life and you get a scrawny guy who kind of dabbles in wrestling, swimming and 20 other sports too. Now these two wrestle. Who wins? Of course, the specialist, the muscle guy wins. Except that it is not what happens at all. Look, the guy who doubles in many sports, that is the AI that played many games, but one particular game only a bit. And it can beat a specialist that played this game a ton. I mean, what? ### Segment 2 (05:00 - 06:00) [5:00] That is insanity. The short version of this is that after it has trained on more than one game, it was able to play all games better. Isn't that the definition of intelligence? It is able to gather knowledge from somewhere and apply it to somewhere else. And that is exactly what we are getting here. Genie 3 can invent infinite new worlds to learn from and Sema can actually do the learning in them. Then see how smart it can become. I cannot wait to see this. What a time to be alive, man. Scientists at Google Deep Mind think on a different level, not just on a level of one work. And man, I have to say I just want to get lost in these incredible worlds right now. Now, the algorithm doesn't really recommend this kind of content anymore. So, I would like to ask you fellow scholars to please leave a kind comment. That is the only way to get these videos recommended to you and others so we can make more. And I got to say I can't stop playing with OpenAI's Open GPT model through Lambda GPU cloud. And as you see I am doing very useful things with it for science. Yes, this is actual speed. I can't believe that I can have more than 100 billion parameters running super fast here. Many of you fellow scholars are using it, and if you don't, make sure to check it out. It costs only a couple dollars per hour. Insanity. You can rent an Nvidia GPU through lambda. ai/papers AI/papers or click the link in the video description.